From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from szxga02-in.huawei.com ([119.145.14.65]:42202 "EHLO szxga02-in.huawei.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1756782Ab2FUBnr (ORCPT ); Wed, 20 Jun 2012 21:43:47 -0400 Message-ID: <4FE27ABB.8060206@huawei.com> Date: Thu, 21 Jun 2012 09:36:59 +0800 From: Jiang Liu MIME-Version: 1.0 To: Yinghai Lu CC: Bjorn Helgaas , Taku Izumi , Kenji Kaneshige , Don Dutile , Yijing Wang , Keping Chen , , Jiang Liu Subject: Re: [PATCH v8 06/12] PCI/ACPI: provide MCFG address for PCI host bridges References: <1340111732-6276-1-git-send-email-jiang.liu@huawei.com> <1340111732-6276-7-git-send-email-jiang.liu@huawei.com> <4FE18A9B.9040501@huawei.com> In-Reply-To: Content-Type: text/plain; charset="ISO-8859-1" Sender: linux-pci-owner@vger.kernel.org List-ID: Hi Yinghai, OK, I will send out v9 without the MCFG cache. Thanks! Gerry On 2012-6-21 2:03, Yinghai Lu wrote: > On Wed, Jun 20, 2012 at 1:32 AM, Jiang Liu wrote: >> Hi Yinghai, >> >> Yes, I'm going to adopt solution two as you suggested. >> On the other hand, the MMCFG caching is kept due to following >> considerations: >> 1) To emit a warning message if MMCFG entries in MCFG table only >> partially covers buses under a PCI host bridge. Taku reported that >> he has a system which exhibits such a behavior. > Interesting. > that should have one overal checking after _CBA entry is added into > pci_mmcfg_list. > We limit root bus bus range after busn_alloc is there. > aka dump the bus range above mmcfg. > or user need to disable mmcfg. > > >> 2) To cross-check that MMCFG addresses returned by MCFG table and >> _CBA method are consistent if both are available (though that >> violates the PCI FW/ACPI specifications). > > now we are adding support for pci hostbridge plug. so we should better to stick > with spec instead of trying to workaround possible FW problem. > Do spoil them too much. > > my point is: MCFG is static. so the range from MCFG can not be changed after > they pass the sanity checking. > later if _CBA is trying come again with overlapping, just through that away. > then if _CBA is good, then just record the range, and later release the range > according the storage during hostbridge removal. > >> 3) In future, we may try to remove MMCFG entry constructed from >> MCFG table when hot-removing a PCI host bridge. We have some systems >> which assign a distinguish segment ID for each host bridge. In such >> a case, it may be reasonable to remove the MMCFG entry when removing >> a host bridge. > No, MCFG is static one. > >> 4) The MCFG cache should be small under normal cases. >> >> If you feel it's unnecessary to keep the cache, I will remove >> it and send out a updated version soon. > for pci host bridge support, we would touch too much thing, I would > like to limit > first round change and keep it simple, and later could optimize it if possible. > > Thanks > > Yinghai > > . >